Credit
We at TERN acknowledge the Traditional Owners and Custodians throughout Australia, New Zealand and all nations. We honour their profound connections to land, water, biodiversity and culture and pay our respects to their Elders past, present and emerging.
This dataset was produced using data sourced from the US Geological Survey.
Purpose
We realised that there was no easily accessible map of woody-vegetation cover of Australia, produced consistently across the continent, for land managers and ecologists to use at a local-scale. Researchers and governments have opened access to their field, airborne and satellite image data, making the task of creating such a map possible. We built on these efforts to create a map of woody-vegetation cover of Australia for the decade from 2000 to 2010.
Lineage
Data Creation
Data:
The foliage projective cover product is derived from an inter-annual time series of the green layer of the Landsat fractional cover product. The Landsat fractional cover product provides an estimates of the vertically-projected fraction of green vegetation, not-green vegetation and bare ground for each pixel. Landsat 5 TM and Landsat 7 ETM+ images were obtained for 374 world wide reference system 2 (WRS2) scenes covering Australia. One dry-season image per year was acquired between 2000 and 2010 for each scene except those where cloud or wet conditions precluded image acquistion for a year. The imagery were processed to BRDF and topographically adjusted reflectance; fractional cover estimates produced; and masks for cloud, cloud shadow, water, topographic shadow, incidence and exitance angle greater than 80 degrees, and snow created.
Statistics:
A robust regression of the form Y~b0 + b1*X, where Y is the green fraction and X is time, was fit to the masked time-series of green vegetation fractions. The following statistics were derived from the regression modelling for each pixel:
1) fitted fraction from the model at 30 June 2005. 2) number of observations in the time series 3) minimum green fraction in the time series once outliers are removed, where an outlier is defined as a point whose residual (observed-fitted) is greater than MAD/0.6745 where MAD is the median absolute deviation of observations from the fitted line. 4) a measure of the standard error of the robust regression fit calculated as sqrt( chisqd/(N-2) ) where N is the number of observations in the time series and chisqd is the weighted sum of squares of residuals. 5) a measure of the normalised standard error of the robust regression fit calculated as standard error divided by the minimum.
Statistics 2:
6) The slope of the regression line in units of percent green fraction per day. 7) The standard deviation of negative residuals, i.e. those observations below the fitted line.
A random forest classifier, using the minimum fraction and standard error was trained on 6597 field or image interpreted observations of woody vegetation presence or absence. The woody foliage projective cover was calculated using P = F - (A*V*tanh(B-F)). F is the robust-regression fitted fraction on 30 June 2005. V is the standard deviation of the negative residuals. A and B were parameters that were optimised and were A=7.93 and B=0.66.
foliage projective cover (dma):
0 - null pixels
100-200 - scaled foliage projective cover. To convert to units use: cover_fraction = pixel*0.01 - 1.
forest cover (dm7):
0 - null pixels
100 - not forest
110-200 - scaled foliage projective cover. To convert to units use: cover_fraction = pixel*0.01 - 1.
Accuracy classes for persistent-green extent (dmb):
0 - null pixels
1 - other wooded lands. That is, classified as woody with a foliage projective cover < 0.1
2 - not woody and a foliage projective cover < 0.1
3 - forest. That is, classified as woody with a foliage projective cover >= 0.1
4 - not woody and fpc >= 0.1.
Accuracy:
The user's and producer's accuracies, respectively, for each class are:
1 - 72.9% and 79.8% [40.4% and 100% after these pixels were reclassified to not persistent-green because their cover fractions were less than 0.1]
2 - 65.4% and 56.3%
3 - 92.2% and 95.5%
4 - 75.7% and 61.3%
Authors
Point of Contact
Publisher
Data Quality